Add DeepSeek V4 target-only CPU inference

.github/workflows/check.yml

@@ -9,9 +9,21 @@ on:
   pull_request:
     branches: [main, dev]
 
+permissions:
+  contents: read
+
+concurrency:
+  group: check-${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+env:
+  PYTHONUTF8: '1'
+
 jobs:
   linux:
+    name: Linux
     runs-on: ubuntu-latest
+    timeout-minutes: 15
     steps:
       - uses: actions/checkout@v4
       - name: make check
@@ -20,7 +32,9 @@ jobs:
   windows:
     # The job that would have caught #68/#137 pre-merge: native MinGW-w64
     # (MSYS2/UCRT64), the exact toolchain the README's Windows port targets.
+    name: Windows UCRT64
     runs-on: windows-latest
+    timeout-minutes: 20
     defaults:
       run:
         shell: msys2 {0}
@@ -45,13 +59,31 @@ jobs:
           name: colibri-windows-exe
           path: c/colibri.exe
 
+  v4-tiny:
+    name: Linux (generated tiny V4 target oracle)
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.13'
+          cache: pip
+          cache-dependency-path: c/tools/requirements-deepseek-v4-tiny.txt
+      - name: Install pinned tiny-fixture dependencies
+        run: python -m pip install -r c/tools/requirements-deepseek-v4-tiny.txt
+      - name: Generate fixture and run target oracle
+        run: make -C c deepseek-v4-tiny-check
+
   macos:
     # clang; libomp for the threaded path (Makefile falls back to
     # single-threaded automatically if it's ever missing).
+    name: macOS (colibri + V4 platform gate)
     runs-on: macos-latest
+    timeout-minutes: 15
     steps:
       - uses: actions/checkout@v4
       - name: install libomp
         run: brew install libomp
-      - name: make check
+      - name: make check (V4 execution gated off)
         run: make -C c check

Makefile

@@ -1,4 +1,4 @@
-.PHONY: all glm portable test check cuda-test clean install uninstall
+.PHONY: all glm deepseek-v4 portable test check cuda-test clean install uninstall
 
-all glm portable test check cuda-test clean install uninstall:
+all glm deepseek-v4 portable test check cuda-test clean install uninstall:
 	$(MAKE) -C c $@

README.md

@@ -457,6 +457,25 @@ Two things that differ per model, both documented in the per-model page:
 | Grammar-forced drafts (structured output) | [docs/grammar-draft.md](docs/grammar-draft.md) |
 | Environment variable inventory | [docs/ENVIRONMENT.md](docs/ENVIRONMENT.md) |
 
+## DeepSeek V4
+
+The experimental CPU path for **DeepSeek V4 Flash** uses native FP4 experts,
+automatic RAM planning, shared `st.h` / `quant.h` infrastructure, and a
+persistent target engine. DSpark is intentionally kept for a separate stacked
+follow-up. The target engine is supported on x86-64/aarch64 Linux and
+Windows/MSYS2.
+
+```bash
+cd c
+make deepseek-v4
+python ./coli run --model /path/to/DeepSeek-V4-Flash --ram 32 \
+  "What is the capital of France?"
+# The same model also works with: coli chat / coli serve / coli web
+```
+
+See [docs/deepseek-v4.md](docs/deepseek-v4.md) for status, checkpoint
+validation, unified CLI/server usage, and the generated tiny independent oracle.
+
 ## What's next
 
 - **Inference-systems research is the product.** The current hierarchy is LRU +
@@ -499,7 +518,7 @@ c/
 └── tests/                dependency-free C and Python tests
 web/                      browser UI (pure OpenAI-API client)
 desktop/                  Tauri v2 desktop shell wrapping the web UI
-docs/                     reference docs, experiments, media
+docs/                     reference docs, experiments, media, DeepSeek V4
 ```
 
 The runtime path intentionally stays flat and readable: `glm.c` plus its small

README.zh-CN.md

@@ -283,6 +283,25 @@ COLI_MODEL=/nvme/glm52_i4 ./coli doctor   # 只读就绪检查
 | 语法强制草稿（结构化输出） | [docs/grammar-draft.md](docs/grammar-draft.md) |
 | 环境变量完整清单 | [docs/ENVIRONMENT.md](docs/ENVIRONMENT.md) |
 
+## DeepSeek V4
+
+实验性的 **DeepSeek V4 Flash** CPU 路径使用原生 FP4 专家、自动 RAM
+规划、共享的 `st.h`／`quant.h` 基础设施，以及常驻 target engine。
+DSpark 有意保留给紧随其后的独立 stacked PR。本 target engine 支持
+x86-64／aarch64 Linux 与 Windows／MSYS2。
+
+```bash
+cd c
+make deepseek-v4
+python ./coli run --model /path/to/DeepSeek-V4-Flash --ram 32 \
+  "法国的首都是哪里？"
+# 同一模型也支持：coli chat / coli serve / coli web
+```
+
+状态、checkpoint 验证、统一 CLI／server 用法和动态生成的 tiny 独立
+oracle 说明，请参阅[中文版 DeepSeek V4 文档](docs/deepseek-v4.zh-CN.md)；
+英文原文见 [docs/deepseek-v4.md](docs/deepseek-v4.md)。
+
 ## 下一步
 
 - **推理系统研究就是产品。**当前层级采用 LRU 与学习型固定集；正在研究模型格式、压缩、
@@ -323,7 +342,7 @@ c/
 └── tests/                零依赖的 C 与 Python 测试
 web/                      浏览器 UI（纯 OpenAI API client）
 desktop/                  封装网页 UI 的 Tauri v2 桌面 shell
-docs/                     参考文档、实验与媒体文件
+docs/                     参考文档、实验、媒体文件与 DeepSeek V4 说明
 ```
 
 运行时路径刻意保持扁平、易读：`colibri.c` 加上若干头文件。

c/Makefile

@@ -250,6 +250,23 @@ else
 PYTHON    ?= python3
 endif
 CUDA_OBJ  =
+
+# DeepSeek V4 is a separate x86-64/aarch64 runtime. Portable infrastructure
+# tests run everywhere; the engine and token-exact fixture run on x86-64
+# Linux/Windows and aarch64 Linux. SIMD fast paths are gated per-feature
+# inside the sources, so a plain armv8-a build stays correct.
+COLI_V4_SUPPORTED :=
+ifneq (,$(X86_64))
+ifneq (,$(IS_WIN)$(LINUX))
+COLI_V4_SUPPORTED := 1
+endif
+endif
+ifneq (,$(AARCH64))
+ifneq (,$(LINUX))
+COLI_V4_SUPPORTED := 1
+endif
+endif
+
 # Gates are exactly the tests that have a build rule further down -- derived from those
 # rules, so adding a gate means adding your tests/test_*.c and its rule, and nothing else.
 # There is no shared list to conflict on.
@@ -261,12 +278,17 @@ CUDA_OBJ  =
 # work-in-progress test, test_fse, test_tok, test_tok_kimi, test_vk_mxfp4) into gates, and
 # they fail to link. Having a rule is the honest definition of "this is a gate".
 TEST_RULES  := $(shell sed -n 's|^tests/\(test_[a-z0-9_]*\)\$$(EXE):.*|\1|p' $(firstword $(MAKEFILE_LIST)))
-# test_uring has a rule but is Linux-only; it is appended conditionally just below.
-TEST_EXCLUDE = test_uring
+# test_uring is Linux-only. V4 engine tests are appended below only on supported
+# x86-64 Linux/Windows and aarch64 Linux hosts; the V4 infrastructure tests have
+# unconditional rules and therefore remain portable gates.
+TEST_EXCLUDE = test_uring test_deepseek_v4 test_v4_ownership
 TEST_BINS    = $(addprefix tests/,$(addsuffix $(EXE),$(filter-out $(TEST_EXCLUDE),$(TEST_RULES))))
 ifneq (,$(LINUX))
 TEST_BINS += tests/test_uring$(EXE)
 endif
+ifeq ($(COLI_V4_SUPPORTED),1)
+TEST_BINS += tests/test_deepseek_v4$(EXE) tests/test_v4_ownership$(EXE)
+endif
 
 # Windows CUDA DLL path: host links the loader, NOT cudart.
 ifneq ($(IS_WIN),)
@@ -368,6 +390,47 @@ endif
 
 all: colibri$(EXE)
 
+.PHONY: deepseek-v4 deepseek-v4-oracle deepseek-v4-tiny-generate \
+	deepseek-v4-tiny-check deepseek-v4-clean
+ifeq ($(COLI_V4_SUPPORTED),1)
+deepseek-v4:
+	$(MAKE) -f Makefile.deepseek-v4 ARCH=$(ARCH) deepseek-v4
+
+deepseek-v4-clean:
+	$(MAKE) -f Makefile.deepseek-v4 deepseek-v4-clean
+
+deepseek-v4-tiny-generate:
+	$(PYTHON) tools/make_deepseek_v4_tiny.py \
+		--output ./deepseek_v4_tiny --force
+
+deepseek-v4-tiny-check: deepseek-v4-tiny-generate
+	$(MAKE) -f Makefile.deepseek-v4 deepseek-v4-clean
+	$(MAKE) -f Makefile.deepseek-v4 ARCH=$(if $(strip $(ARCH)),$(ARCH),$(PORTABLE_ARCH)) deepseek-v4
+	$(PYTHON) tests/test_deepseek_v4_tiny.py \
+		--binary ./$(if $(IS_WIN),deepseek_v4.exe,deepseek_v4) \
+		--fixture ./deepseek_v4_tiny
+
+deepseek-v4-oracle: deepseek-v4
+	@test -n "$(MODEL)" || { echo "usage: make deepseek-v4-oracle MODEL=/path/to/checkpoint" >&2; exit 2; }
+	$(PYTHON) tools/make_deepseek_v4_oracle.py --model "$(MODEL)" \
+		--binary ./$(if $(IS_WIN),deepseek_v4.exe,deepseek_v4) \
+		--output tests/deepseek_v4_oracle.json \
+		--memory-gb $(or $(MEMORY_GB),32) \
+		--prompt "$(or $(ORACLE_PROMPT),The capital of France is)" \
+		--validate \
+		--teacher-forcing $(or $(ORACLE_TEACHER_FORCING),32) \
+		--greedy $(or $(ORACLE_GREEDY),20)
+else
+deepseek-v4 deepseek-v4-oracle:
+	@echo "$@ is supported only on x86-64/aarch64 Linux and Windows/MSYS2" >&2; exit 1
+deepseek-v4-clean:
+	@:
+deepseek-v4-tiny-check:
+	@echo "SKIP deepseek-v4-tiny-check: V4 runtime requires x86-64/aarch64 Linux or Windows/MSYS2"
+deepseek-v4-tiny-generate:
+	@echo "SKIP deepseek-v4-tiny-generate: V4 runtime requires x86-64/aarch64 Linux or Windows/MSYS2"
+endif
+
 # phony targets — 'glm' kept for backward compatibility
 colibri: colibri$(EXE)
 glm: colibri$(EXE)
@@ -553,6 +616,9 @@ tests/test_tok_o200k$(EXE): tests/test_tok_o200k.c tok.h tok_unicode.h tok_unico
 tests/test_st_pread$(EXE): tests/test_st_pread.c st.h json.h compat.h
 	$(CC) $(CFLAGS) -DST_PREAD_CHUNK=7 $< -o $@ $(LDFLAGS)
 
+tests/test_st_shape$(EXE): tests/test_st_shape.c st.h json.h compat.h
+	$(CC) $(CFLAGS) $< -o $@ $(LDFLAGS)
+
 tests/test_st$(EXE): tests/test_st.c st.h json.h compat.h
 	$(CC) $(CFLAGS) $< -o $@ $(LDFLAGS)
 
@@ -649,6 +715,59 @@ tests/test_i4_acc512$(EXE): tests/test_i4_acc512.c
 tests/test_compat_direct$(EXE): tests/test_compat_direct.c compat.h
 	$(CC) $(CFLAGS) $< -o $@ $(LDFLAGS)
 
+tests/test_expert_store_ops$(EXE): tests/test_expert_store_ops.c expert_store.h tensor.h
+	$(CC) $(CFLAGS) $< -o $@ $(LDFLAGS)
+
+tests/test_native_quant$(EXE): tests/test_native_quant.c deepseek_v4.c native_quant.h tensor.h quant.h
+	$(CC) $(CFLAGS) -DCOLI_V4_UNIT_NATIVE_QUANT deepseek_v4.c $< -o $@ $(LDFLAGS)
+
+ifeq ($(COLI_V4_SUPPORTED),1)
+include Makefile.deepseek-v4.units
+V4_TEST_LINK_OBJS = \
+	COLI_V4_UNIT_ST.o \
+	COLI_V4_UNIT_CONFIG.o \
+	COLI_V4_UNIT_MATH.o \
+	COLI_V4_UNIT_SPARSE_ATTENTION.o \
+	COLI_V4_UNIT_RESOURCE_PLAN.o \
+	COLI_V4_UNIT_PROMPT.o \
+	$(addsuffix .o,$(V4_TEST_UNITS)) \
+	COLI_V4_UNIT_NATIVE_QUANT.o
+
+tests/test_deepseek_v4$(EXE): tests/test_deepseek_v4.c deepseek_v4.c deepseek_v4.h \
+		Makefile.deepseek-v4.units
+	$(MAKE) -f Makefile.deepseek-v4 ARCH=$(ARCH) deepseek-v4-test-objs \
+		COLI_V4_UNIT_ST.o COLI_V4_UNIT_CONFIG.o COLI_V4_UNIT_MATH.o COLI_V4_UNIT_SPARSE_ATTENTION.o \
+		COLI_V4_UNIT_RESOURCE_PLAN.o COLI_V4_UNIT_PROMPT.o COLI_V4_UNIT_NATIVE_QUANT.o
+	$(CC) $(CFLAGS) $< $(V4_TEST_LINK_OBJS) -o $@ -pthread $(LDFLAGS)
+
+V4_OWN_DIR = build/ownership
+V4_OWN_CFLAGS = $(CFLAGS) -DCOLI_V4_TEST_HOOKS
+V4_OWNERSHIP_OBJS = \
+	$(V4_OWN_DIR)/COLI_V4_UNIT_RUNTIME.o \
+	$(V4_OWN_DIR)/COLI_V4_UNIT_CONFIG.o \
+	$(V4_OWN_DIR)/COLI_V4_UNIT_ST.o \
+	$(V4_OWN_DIR)/COLI_V4_UNIT_NATIVE_QUANT.o
+
+$(V4_OWN_DIR):
+	mkdir -p $(V4_OWN_DIR)
+
+$(V4_OWN_DIR)/COLI_V4_UNIT_RUNTIME.o: deepseek_v4.c deepseek_v4.h deepseek_v4_internal.h | $(V4_OWN_DIR)
+	$(CC) $(V4_OWN_CFLAGS) -DCOLI_V4_UNIT_RUNTIME -c deepseek_v4.c -o $@
+
+$(V4_OWN_DIR)/COLI_V4_UNIT_CONFIG.o: deepseek_v4.c deepseek_v4.h deepseek_v4_internal.h | $(V4_OWN_DIR)
+	$(CC) $(V4_OWN_CFLAGS) -DCOLI_V4_UNIT_CONFIG -c deepseek_v4.c -o $@
+
+$(V4_OWN_DIR)/COLI_V4_UNIT_ST.o: deepseek_v4.c deepseek_v4.h deepseek_v4_internal.h st.h | $(V4_OWN_DIR)
+	$(CC) $(V4_OWN_CFLAGS) -DCOLI_V4_UNIT_ST -c deepseek_v4.c -o $@
+
+$(V4_OWN_DIR)/COLI_V4_UNIT_NATIVE_QUANT.o: deepseek_v4.c deepseek_v4.h deepseek_v4_internal.h quant.h | $(V4_OWN_DIR)
+	$(CC) $(V4_OWN_CFLAGS) -DCOLI_V4_UNIT_NATIVE_QUANT -c deepseek_v4.c -o $@
+
+tests/test_v4_ownership$(EXE): tests/test_v4_ownership.c $(V4_OWNERSHIP_OBJS)
+	$(CC) $(V4_OWN_CFLAGS) $< $(V4_OWNERSHIP_OBJS) -o $@ -pthread $(LDFLAGS)
+
+endif
+
 tests/test_route_trace$(EXE): tests/test_route_trace.c route_trace.h
 	$(CC) $(CFLAGS) $< -o $@ $(LDFLAGS)
 
@@ -733,13 +852,16 @@ check:
 	$(MAKE) portable
 	$(MAKE) test
 
-install: colibri$(EXE) olmoe$(EXE)
+install: colibri$(EXE) olmoe$(EXE) $(if $(filter 1,$(COLI_V4_SUPPORTED)),deepseek-v4,)
 	$(INSTALL) -d $(DESTDIR)$(BINDIR)
 	$(INSTALL) -d $(DESTDIR)$(LIBEXECDIR)
 	$(INSTALL) -d $(DESTDIR)$(LIBEXECDIR)/tools
 	$(INSTALL) -m 755 coli $(DESTDIR)$(BINDIR)/coli
 	$(INSTALL) -m 755 colibri$(EXE) $(DESTDIR)$(LIBEXECDIR)/colibri$(EXE)
 	$(INSTALL) -m 755 olmoe$(EXE) $(DESTDIR)$(LIBEXECDIR)/olmoe$(EXE)
+	@if [ -f deepseek_v4$(EXE) ]; then \
+	  $(INSTALL) -m 755 deepseek_v4$(EXE) $(DESTDIR)$(LIBEXECDIR)/deepseek_v4$(EXE); \
+	fi
 	$(INSTALL) -m 644 resource_plan.py doctor.py autotune.py openai_server.py version.py $(DESTDIR)$(LIBEXECDIR)/
 	$(INSTALL) -m 644 tools/*.py $(DESTDIR)$(LIBEXECDIR)/tools/
 	@# The dashboard is an optional build artifact (cd web && npm run build), so install
@@ -761,7 +883,10 @@ clean:
 
 bench: iobench$(EXE)
 	@if [ -n "$(ARGS)" ]; then ./iobench$(EXE) $(ARGS); else echo "built iobench$(EXE) — run: ./iobench$(EXE) <file> <MB> <iters> <threads> <direct 0|1>"; fi
-.PHONY: all colibri glm iq3 rans fuzz-rans cuda-test hip-test gpu-compile cuda-bench cuda-dll portable test-c test-python test check clean install uninstall bench
+.PHONY: all colibri glm deepseek-v4 deepseek-v4-clean deepseek-v4-oracle \
+	deepseek-v4-tiny-generate deepseek-v4-tiny-check \
+	iq3 rans fuzz-rans cuda-test hip-test gpu-compile cuda-bench cuda-dll portable \
+	test-c test-python test check clean install uninstall bench
 
 tests/test_kv_prefix$(EXE): tests/test_kv_prefix.c kv_prefix.h
 	$(CC) $(CFLAGS) tests/test_kv_prefix.c -o tests/test_kv_prefix$(EXE) $(LDFLAGS)